Patent Grant
8065130
One or more aspects of the present invention relate generally to integrated circuit design tools and, more particularly, to a programmable architecture for implementing a message processing system using an integrated circuit.
Programmable logic devices (PLDs) exist as a well-known type of integrated circuit (IC) that may be programmed by a user to perform specified logic functions. There are different types of programmable logic devices, such as programmable logic arrays (PLAs) and complex programmable logic devices (CPLDs). One type of programmable logic device, known as a field programmable gate array (FPGA), is very popular because of a superior combination of capacity, flexibility, time-to-market, and cost.
An FPGA typically includes an array of configurable logic blocks (CLBs) surrounded by a ring of programmable input/output blocks (IOBs). The CLBs and IOBs are interconnected by a programmable interconnect structure. The CLBs, IOBs, and interconnect structure are typically programmed by loading a stream of configuration data (known as a bitstream) into internal configuration memory cells that define how the CLBs, IOBs, and interconnect structure are configured. Additionally, an FPGA may include embedded memory, such as block random access memories (BRAMs), one or more microprocessors, sometimes referred to as embedded cores, and digital clock managers (DCMs). The combination of components on an FPGA may be used for system-level integration, sometimes referred to as “system-on-a-chip” (SOC).
Historically, FGPAs have not been employed in network processing applications. Rather, Network devices, such as routers, employ dedicated, special purpose components for processing packets that propagate through the network. Conventionally, network devices employ network processors or application specific integrated circuits (ASICs) to provide the desirable packet processing/network processing functions. Such processor- or ASIC-based architectures, however, are static in nature, providing a fixed amount of resources for packet processing/network processing functions. Accordingly, there exists a need in the art for more flexible message processing architectures.
One aspect of the invention relates to message processing using an integrated circuit. In one embodiment, configurable logic of an integrated circuit is configured to have a plurality of thread circuits and a memory. Messages are received to the integrated circuit for storage in the memory. The memory is accessed with the plurality of threads to concurrently process a plurality of the messages.
Another aspect of the invention relates to designing a message processing system for implementation using an integrated circuit. Specification data is received for specifying processing operations and memory attributes of the message processing system. A logical description of the message processing system is generated in response to the specification data, where the logical description includes a memory component and a processing component. A physical description of the message processing system is generated in response to the logical description. The physical description includes at least one memory circuit for implementing the memory component and a plurality of threads for implementing the processing component.
Another aspect of the invention relates to a design tool for designing a message processing system for implementation using an integrated circuit. An input section is configured to specify processing operations and memory attributes of the message processing system. A first database is configured to store a message processing architecture having a memory component and a processing component. A second database is configured to store an integrated circuit architecture. A message processing section includes a first portion and a second portion. The first portion is configured to generate an instance of the memory component and an instance of the processing component to produce a logical view of the message processing system. A second portion is configured to implement the memory component instance and the processing component instance in terms of the integrated circuit architecture to produce a physical view of the message processing system.
Another aspect of the invention relates to message processing using an integrated circuit. Circuitry is configured to receiving message to the integrated circuit. A memory is configured to storing the messages. A first configured portion within the integrated circuit is adapted for concurrently accessing the memory with a plurality of thread circuits to process the messages.
Accompanying drawing(s) show exemplary embodiment(s) in accordance with one or more aspects of the invention; however, the accompanying drawing(s) should not be taken to limit the invention to the embodiment(s) shown, but are for explanation and understanding only.
To facilitate understanding of the invention, the description has been organized as follows:
Overview, introduces aspects of the invention and exemplary embodiments of their relationships to one another;
Soft Platform, describes a programmable architecture and associated design tool for implementing a message processing system using an integrated circuit;
Memory Model, describes an application-specific programmable memory architecture and interconnection network for an integrated circuit;
Multithread Model, describes an inter-process synchronization mechanism for threads implemented within a configurable logic portion of an integrated circuit; and
Programming Interface, describes a programming interface for a design tool embodying a soft architecture for implementing a message processing system using an integrated circuit.
Overview
One or more aspects of the invention are related to a configurable and programmable micro-architecture for implementing message-processing (MP) systems (“soft platform architecture”). As used herein, the term “message” encompasses packets, cells, frames, data units, and like type blocks of information known in the art that is passed over a communication channel. A “message-processing” system is a system or subsystem for processing messages (e.g., a packet processing system or a network processing system). The soft platform architecture is “message-centric” to match the nature of MP systems. That is, the processing components of the MP system go to the messages, as opposed to the messages coming to the processing components.
Briefly stated, a designer specifies attributes for an MP system, such as structural and behavioral attributes for processing components and memory components. For example, the designer may employ a set of descriptions or “primitives” that parametrically define the MP system attributes. The primitives provide an abstract mechanism for defining the MP system. A design tool embodying the soft platform architecture may include a programming interface for generating a logical description or “logical view” of an MP system based on the designer-specified attributes.
Notably, the logical view includes logical components of the soft platform architecture configured in accordance with the designer-specified MP system. In particular, the soft platform architecture includes a memory model component and a multithreading component. A physical view of the MP system may then be generated based on the logical view. The physical view includes physical components of an integrated circuit architecture that implement the logical components of the soft platform architecture. The physical view may then be processed to generate configuration data for the integrated circuit to realize the designer-specified MP system (e.g., a configuration bitstream for a PLD or mask data for an ASIC). Thus, the soft platform architecture provides a mechanism by which a designer may design an MP system in an abstract fashion, without knowledge of the particular physical configuration of the integrated circuit.
One or more aspects of the invention are described with respect to a programmable architecture for implementing a message processing system using an FPGA. While the invention is described with specific reference to an FPGA, those skilled in the art will appreciate that other types of programmable logic devices may be used, such as complex programmable logic devices (CPLDs). In addition, other types of mask-programmable devices may be used, such as application specific integrated circuits (ASICs). Those skilled in the art will appreciate that, if an ASIC is employed rather than an PLD, then mask data is generated in place of a configuration bitstream.
The FPGA 102 illustratively comprises programmable logic circuits or “blocks”, illustratively shown as CLBs 104, IOBs 106, and programmable interconnect 108 (also referred to as “programmable logic”), as well as configuration memory 116 for determining the functionality of the FPGA 102. The FPGA 102 may also include an embedded processor block 114, as well as various dedicated internal logic circuits, illustratively shown as blocks of random access memory (“BRAM 110”), configuration logic 118, digital clock management (DCM) blocks 112, and input/output (I/O) transceiver circuitry 122. Those skilled in the art will appreciate that the FPGA 102 may include other types of logic blocks and circuits in addition to those described herein.
As is well known in the art, the IOBs 106, the CLBs 104, and the programmable interconnect 108 may be configured to perform a variety of functions. Notably, the CLBs 104 are programmably connectable to each other, and to the IOBs 106, via the programmable interconnect 108. Each of the CLBs 104 may include one or more “slices” and programmable interconnect circuitry (not shown). Each CLB slice in turn includes various circuits, such as flip-flops, function generators (e.g., a look-up tables (LUTs)), logic gates, memory, and like type well-known circuits. The IOBs 106 are configured to provide input to, and receive output from, the CLBs 104.
Configuration information for the CLBs 104, the IOBs 106, and the programmable interconnect 108 is stored in the configuration memory 116. The configuration memory 116 may include static random access memory (SRAM) cells. The configuration logic 118 provides an interface to, and controls configuration of, the configuration memory 116. A configuration bitstream produced from the program memory 120 may be coupled to the configuration logic 118 through a configuration port 119. The configuration process of FPGA 102 is also well known in the art.
The I/O transceiver circuitry 122 may be configured for communication over any of a variety of media, such as wired, wireless, and photonic, whether analog or digital. The I/O transceiver circuitry 122 may comprise gigabit or multi-gigabit transceivers (MGTs). The DCM blocks 112 provide well-known clock management circuits for managing clock signals within the FPGA 102, such as delay lock loop (DLL) circuits and multiply/divide/de-skew clock circuits.
The processor block 114 comprises a microprocessor core, as well as associated control logic. Notably, such a microprocessor core may include embedded hardware or embedded firmware or a combination thereof for a “hard” or “soft” microprocessor. A soft microprocessor may be implemented using the programmable logic of the FPGA 102 (e.g., CLBs 104, IOBs 106). For example, a Microblaze™ soft microprocessor, available from Xilinx® of San Jose, Calif., may be employed. A hard microprocessor may be implemented using an IBM Power PC, Intel Pentium, AMD Athlon, or like type processor core known in the art.
The processor block 114 is coupled to the programmable logic of the FPGA 102 in a well known manner. For purposes of clarity by example, the FPGA 102 is illustrated with 12 CLBs, 16 IOBs, 4 BRAMs, 4 DCMs, and one processor block. Those skilled in the art will appreciate that actual FPGAs may include one or more of such components in any number of different ratios. For example, the FPGA 102 may be selected from the Virtex™-II Pro family of products, commercially available from Xilinx® of San Jose, Calif.
One or more aspects of the invention include design tools for designing MP systems, memory systems, and multithreading systems. Such design tools may be implemented using a computer. Notably,
The memory 1703 may store all or portions of one or more programs and/or data to implement the processes, methods, and design tool sections described herein. Although one or more aspects of the invention are disclosed as being implemented as a computer executing a software program, those skilled in the art will appreciate that the invention may be implemented in hardware, software, or a combination of hardware and software. Such implementations may include a number of processors independently executing various programs and dedicated hardware, such as ASICs.
The computer 1700 may be programmed with an operating system, which may be OS/2, Java Virtual Machine, Linux, Solaris, Unix, Windows, Windows95, Windows98, Windows NT, and Windows2000, WindowsME, and WindowsXP, among other known platforms. At least a portion of an operating system may be disposed in the memory 1703. The memory 1703 may include one or more of the following random access memory, read only memory, magneto-resistive read/write memory, optical read/write memory, cache memory, magnetic read/write memory, and the like, as well as signal-bearing media as described below.
An aspect of the invention is implemented as a program product for use with a computer system. Program(s) of the program product defines functions of embodiments and can be contained on a variety of signal-bearing media, which include, but are not limited to: (i) information permanently stored on non-writable storage media (e.g., read-only memory devices within a computer such as CD-ROM or DVD-ROM disks readable by a CD-ROM drive or a DVD drive); (ii) alterable information stored on writable storage media (e.g., floppy disks within a diskette drive or hard-disk drive or read/writable CD or read/writable DVD); or (iii) information conveyed to a computer by a communications medium, such as through a computer or telephone network, including wireless communications. The latter embodiment specifically includes information downloaded from the Internet and other networks. Such signal-bearing media, when carrying computer-readable instructions that direct functions of the invention, represent embodiments of the invention.
Soft Platform
Notably, the FPGA circuit design may be “interface-centric” in that the circuit design is driven by the behavior at the system interfaces, as opposed to the “processor-centric” model, where the circuit design is driven by the behavior of an embedded processor. The interface-centric circuit design model matches well with the message-centric style of the soft platform architecture. Placement and usage of interfaces, memories, and their interconnections dominate the allocation of FPGA architecture features, and then allocation of functional elements (e.g., programmable logic, embedded processors) for the process components follow as a derivative.
More specifically, the input section 202 is coupled to one or more input devices 210 and a database storing an application programming interface (API) (“API database 212”). The API database 212 includes a set of primitives associated with structural and behavioral attributes of the soft platform architecture. Thus, the API provides a “programming interface” for the soft platform architecture. An exemplary embodiment of a programming interface for a soft platform architecture is described below in the section entitled “PROGRAMMING INTERFACE.” Using the input devices 210, a designer may interact with the input section 202 to produce specification data for an MP system or subsystem if the MP circuit is part of a larger system (hereinafter referred to as an “MP system”).
Notably, in one embodiment, a designer may use the primitives in the API database 212 directly to produce the MP system specification data for the soft platform architecture. In another embodiment, a designer may design the MP system using alternate constructions provided by the input section 202. That is, the input section 202 may comprise a design entry tool specific to the MP domain. Examples of such MP-specific design-entry tools include Click (available from The Massachusetts Institute of Technology), Rave (available from Cloudshield™ of Sunnyvale, Calif.), and SDL (a telecom standard from ITU-T). The input section 202 may then map the MP system specified using the alternate constructions onto the primitives in the API database 212 for the soft platform architecture. Thus, the input section 202 may provide a different level of abstraction than that provided by the soft platform architecture.
In one embodiment of the invention, the MP system specification data may comprise program code for programmatically interacting with the soft platform section 204. The program code may be callable by an external design tool of the input section 202. In another embodiment, the MP system specification data may comprise interpretive descriptions (e.g., descriptions in a textual or binary format) that the soft platform section 204 may interpret (e.g., an XML format). In either embodiment, the MP system specification is used to configure the soft platform architecture.
The soft platform section 204 is coupled to the input section 202 for receiving the MP system specification data. The soft platform section 204 is also coupled to a database storing the features or attributes of the soft platform architecture (“soft platform database 216”), and a database storing features or attributes of the architecture of the FPGA 208 (“FPGA database 218”).
The soft platform section 204 includes a first portion 203 for generating a logical description or “logical view” of an MP system in accordance with the MP system specification. The logical view is defined in terms of the logical components of the soft platform architecture stored in the soft platform database 216. The soft platform section 204 includes a second portion 205 for generating a physical view of the MP system. Notably, using information in the FPGA database 218, the soft platform section 204 maps the logical view of the MP system defined in terms of the soft platform architecture onto the architecture of the FPGA 208. The soft platform section 204 provides FPGA design data as output, which represents a “physical view” of the MP system in terms of the architecture of the FPGA 208. Details of the soft platform architecture are described below with respect to
The FPGA design tools section 206 is coupled to the soft platform section 204 for receiving the FPGA design data. The FPGA design data may comprise a physical description of the MP system specified by the designer in terms of the components and features of the FPGA 208. For example, in one embodiment, the FPGA design data may comprise a hardware description language (HDL) representation of the MP system design (e.g., Very high-speed integrated circuit description language (VHDL) or Verilog). The FPGA design tools section 206 processes the FPGA design data to produce configuration bitstream data. For example, the FPGA design tools section 206 may comprise various well-known FPGA design tools, such as a synthesis tool, a map/place/route tool, like-type tools known in the art. The FPGA design tools section 206 provides configuration bitstream data as output, which may be loaded into the FGPA 208.
At step 306, a logical view of the MP system is generated in accordance with the MP system specification. As described above, the logical view of the MP system is defined in terms of a soft platform architecture. The logical components of the soft platform architecture are configured in accordance with the MP system specification to generate the logical view of the MP system. The term “logical components” refers to both the structural and behavioral attributes of the soft platform architecture, described in more detail below.
At step 308, the logical view of the MP system is mapped onto an FPGA architecture to produce FPGA design data. That is, the logical components comprising the logical view are linked to physical components of an FPGA and, optionally, other devices connected to the FPGA (e.g., external memories). In one embodiment of the invention, the FPGA design data comprises an HDL representation of the MP system design. As described above, the FPGA design data provides a physical view of the specified MP system in terms of the architecture of the FPGA. That is, FPGA design data corresponds to the physical implementation of the logical view of the MP system on an FPGA device.
At step 310, the FPGA system design is processed to produce configuration bitstream data. For example, if the FPGA system design comprises an HDL representation of the MP system design, the FPGA system design may be synthesized, mapped, placed, and routed in a well-known manner to produce bitstream data for configuring an FPGA. At step 312, the configuration bitstream data is loaded into an FPGA to realize the MP system specified at step 304. The process 300 ends at step 314.
Notably, the soft platform architecture 400 includes a programming/control interface 414 and a debug/test/monitor interface 416. The programming/control interface 414 conveys the data for configuring the programmable portions of the soft platform architecture 400. The programming/control information conveyed via the programming/control interface 414 comprises the structural and behavioral information related to the MIS component 402, the process component 403, and the SSS component 410. An exemplary embodiment of a programming interface to the soft platform architecture 400 is described below in the section entitled “PROGRAMMING INTERFACE.” The debug/test/monitor interface 416 may be used during the design and implementation of an MP system defined in terms of the soft platform architecture 400. The interfaces 414 and 416 are illustrative, as there may be a single shared interface, or more than two interfaces.
The MIS component 402 is the logical storage point for all messages currently within the system implemented using the soft platform architecture 400. The MIS component 402 includes an interface 412 to the enclosing environment 450 allowing for the input and output of messages. For example, the soft platform architecture 400 may be configured to produce an internet protocol (IP) packet router. The MIS component 402 may be configured to store all IP packets currently in flight through the router. The interface 412 may be one or more ports by which the router is connected to a physical network.
The MIS component 402 may be physically implemented using a centralized memory device, a plurality of distributed memory devices, or a combination thereof. In addition, the types, sizes, and interconnections of the physical memory elements, as well as the interface to such physical memory elements, are programmable through configuration of the MIS component 402. An exemplary embodiment of a logical memory configuration that may be used as the MIS component 402 is described below in the section entitled “MEMORY MODEL.”
The process component 403 comprises one or more processes that may be classified as fine grain operations processes (FOPs) 404, coarse grain operations processes (COPs) 406, or perimeter interface processes (PIPs) 408. In addition, the process component 403 includes an inter-process synchronization component 418. The FOPs 404, COPs 406, and PIPs 408 operate on messages stored within the MIS component 402. The term “process,” as used herein, denotes a concurrent agent for operating on information stored within the MIS component 402. The term “thread” is used to denote an instance of a process.
Notably, each single execution of a process within the process component 403 is associated with a message stored in the MIS component 402 through a message context 405. A process in the process component 403 may be physically implemented directly in programmable logic of an FPGA, or in a soft or hard embedded processor of an FPGA. In one embodiment of the invention, the message context 405 may be implemented using a data counter (DC) component 407. The DC component 407 points to the current position in the current message being operated on by a particular process in the process component 403. As the process executes, the DC component 407 may be updated, either automatically to advance to the next position in a message, or by execution of programmed “data jumps.” In essence, the process moves over the message. The DC component 407 may be physically implemented using a memory element within the FPGA capable of storing an address associated with the location of a message in the memory of the MIS component 402. Depending on the memory organization of the MIS component 402, the DC component 407 may be a register, a BRAM, or an external RAM.
The processes of the process component 403 include a common interface. The inter-process communication component 418 may utilize the common interface to allow interaction between processes of the process component 403. Such interactions may include, for example, creating or destroying a process or passing data to another process. The inter-process communication component 418 provides for a control flow in the processing of a message. At a microscopic level, the inter-process communication component 418 is capable of providing a control flow within a single process's execution. At a macroscopic level, the inter-process communication component 418 is capable of providing a control flow from one process's execution to another process's execution. An exemplary embodiment of a multithread model that may be used as the inter-process communication component 418 is described below in the section entitled “MULTITHREAD MODEL.”
A FOP 404 is the basic programmable unit for message processing. A FOP 404 performs a sequence of steps on a message stored within the MIS component 402. At each step, a set of concurrent operations are performed. A FOP 404 may be associated with a DC component 407. After each step, the DC component 407 may be incremented, or a data jump operation performed, such that the FOP 404 accesses a new portion of the message. The steps, as well as the operations performed during each step, may be programmable, static, or partially programmable and partially static in their definition. Examples of operations include, inspecting a field (e.g., a 16-bit header field) of a message, or performing simple arithmetic (e.g., adding one to a 16-bit header field) on a message.
A FOP 404 may be implemented within an FPGA using programmable logic. For example, a FOP may be implemented as a finite state machine (FSM) configured within the programmable logic of the FPGA. Alternatively, a FOP may be implemented on an embedded processor within an FPGA. For example, a FOP may be implemented as an operating system thread executed by the embedded processor. The physical implementation of a FOP 404 may be programmable, static, or partially programmable and partially static in its definition.
A COP 406 is used to incorporate a function block to perform a message processing operation. A function block may comprise a circuit or subsystem defined outside the context of the soft platform architecture 400. For example, the function block may comprise a reusable intellectual property (IP) core for an FPGA. A COP 406 provides a programmable adapter between the interface of the function block and the common interface of the process component 403. A COP 406 may be started, stopped, or interrupted by another process of the process component 403.
A COP 406 may be defined statically and be in existence permanently. Alternatively, a COP 406 may be created and destroyed dynamically to allow dynamic reconfiguration of the function blocks associated therewith. For example, a COP 406 may be used to incorporate a function block for compression or encryption of all or part of a message stored in the MIS component 402. A COP 406 may be associated with a DC component 407, which points to the beginning of the message in the MIS component 402 to be processed by the COP 406.
A PIP 408 is concerned with enabling the movement of a message to and from soft platform architecture 400. In one embodiment of the invention, a PIP 408 may be used to incorporate a function block, similar to a COP 406. The function block associated with a PIP 408 may comprise a circuit or subsystem defined outside the context of the soft platform architecture 400 that is specifically geared to perform I/O functions. In another embodiment of the invention, a PIP 408 may be implemented as a FSM in programmable logic of the FPGA.
For example, a PIP may be used to receive or transmit successive words of a message over an interface using a protocol defined for the interface. For example, a PIP may act as a smart adapter for the Xilinx® LocalLink interface to a networking core or the interface to a Gigabit MAC core. A PIP may also communicate with other system components implemented within the FPGA.
The SSS component 410 may be used to store state information associated with the processes of the process component 403. For example, the SSS component 410 may be used to store a message context 405 for a FOP 404. The SSS component 410 may be physically implemented using a centralized memory device, a plurality of distributed memory devices, or a combination thereof.
Memory Model
In particular, the input section 502 is coupled to a database that stores an API associated with the memory model, referred to herein as the memory interconnection description language (MIDL) library 506. The MIDL library 506 comprises a set of primitives for defining structural and behavioral attributes of the memory model. Thus, the MIDL library 506 provides a programming interface for the memory model. A designer may interact with the input section 502 to produce specification data for a memory subsystem. The designer may work directly with the MIDL library 506, or may work indirectly with the MIDL library 506 through an alternative design tool defined within the input section 502. The memory subsystem specification data may be programmatic or may be interpretive (e.g., XML). An example of an MIDL specification for a 32-bit wide memory constructed from two 16-bit wide memories, which are in turn mapped to physical BRAM in an FPGA, is shown in Appendix A.
The memory model section 504 is coupled to the input section 502 for receiving the memory model specification. The memory model section 504 is also coupled to a database that stores the features or attributes of the memory model (“memory model database 508”), and a database that stores the memory attributes of an FPGA and external memories associated therewith (“FPGA memory database 510”). The memory model section 504 includes a first portion 503 for generating a logical view of a memory subsystem in accordance with the memory subsystem specification. The logical view is defined in terms of the logical components of the memory model stored in the memory model database 508.
The memory model section 504 may include an analysis/optimization portion 512 for analyzing and optimizing the logical view of the memory subsystem in accordance with constraint data provided by a designer. The memory model section 504 further includes a second portion 505 for generating a physical view of the memory system based on the logical view. Notably, using information in the FPGA memory database 510, the memory model section maps the logical view of the memory subsystem onto the physical memory components associated with an FPGA. The memory model section 504 provides FPGA design data as output.
The memory model 600 is characterized by a “memory architecture” and a “memory-interconnection architecture.” The memory architecture pertains to the size, type, and topology of one or more memory circuits comprising the memory element 602. The memory-interconnection architecture pertains to the type, bus width (e.g., number of wires), and topology of interconnection (e.g., crossbar) of the one or more memory circuits comprising the memory element 602. In general, with respect to the memory model 600, the term “interface” imparts knowledge related to the protocols that must be adhered to for the particular interaction, whereas the term “architecture” imparts knowledge related to the critical path that particular data follows within the memory model 600.
The memory and interconnection architectures of the memory model 600 may be defined by the physical location of the memory circuits used to implement the model, as well as the logical configuration of the interface to such memory circuits. For example, the memory may be physically centralized (i.e., a single physical memory circuit), or several memory circuits may be physically distributed. The memory circuit(s) used to implement the memory model 600 may be disposed within the FPGA (e.g., any combination of on-chip BRAMs, LUT-based RAMs, and shift registers), disposed external to the FPGA (e.g., external SDRAMs, DDR SDRAMs, and RDRAMs), or a combination thereof. In addition, the interface to such memory circuit(s) may be logically centralized (e.g., a unified programming interface) or logically distributed (e.g., multiple logical interfaces).
In light of the various physical and logical configurations for the memory and interconnection architectures, various logical schemes for storing messages may be implemented using the memory model 600. In one embodiment, all messages may be stored within a single memory (e.g., a queue of messages in a memory) (“uniform message storage”). Alternatively, different messages may be allocated over different memories (“interleaved message storage”). In yet another alternative, each message may be physically allocated over different memories (“striped message storage”). In another embodiment, each message may be logically allocated over different memories (“separated message storage”).
In particular,
Notably, an incoming message may be “striped” across the BRAMs 902 such that each of the BRAMs 902 stores only a portion of the incoming message. Each of the computational elements 906 may then access respective ones of the BRAMs 902 through the respective memory interface 904 to access a portion of the incoming message. The memory subsystem 900 is an example of striped message storage using physically distributed memories within an FPGA.
The BRAMs 1002 are logically part of one centralized memory with a dedicated memory interface 1004 that manages access to the messages stored in the BRAMs 1002. Each of the computational elements 1006 may access a message or portion thereof through the memory interface 1004. The memory subsystem 1000 is an example of a logically centralized, physically distributed memory organization.
Returning to
Notably,
The computational element 608 requests data using the control bus 1110. The control logic 1104 determines whether the data is available within the memory 1102. If so, the data is communicated to the computational element 608 over the data bus 1108. Otherwise, the control logic 1104 generates a status signal for communication to the computational element 608 over the status bus 1112. The status signal may indicate the unavailability of the requested data and an estimated duration after which the data will be available.
At step 706, a logical view of the memory subsystem is generated in accordance with the memory subsystem specification. The logical view is defined in terms of the memory model 600. That is, the logical components of the memory model 600 are configured in accordance with the memory subsystem specification to generate a logical view of the memory subsystem.
At step 707, the memory subsystem may be analyzed and optimized in accordance with predefined constraint and test data. The constraint data may include constraints on memory access, time, and interconnect resources. The test data may include one or more test memory access patterns. In one embodiment of the invention, an analysis model based on memory access, time, and interconnect resources is employed.
Notably,
Returning to
At step 710, the FPGA design data may be combined with other FPGA design data to define a system. For example, the memory subsystem may be incorporated into an MP system designed as described above in the section entitled “SOFT PLATFORM.” That is, the memory subsystem may be the implementation of the MIS component of the soft platform architecture used to implement an MP system.
At step 712, the combined FPGA design data is processed to produce configuration bitstream data. For example, if the combined FPGA design data comprises an HDL representation, the FPGA design data may be synthesized, mapped, placed, and routed in a well-known manner to produce bitstream data for configuring an FPGA. At step 716, the configuration bitstream data is loaded into an FPGA. The process 700 ends at step 718.
Multithread Model
In particular, the input section 1202 is coupled to a database that stores a library of multithreading primitives (“multithread primitive database 1206”). The multithread primitive database 1206 stores a set of primitives for defining structural and behavioral attributes of the multithread model. Thus, the multithread primitive database 1206 provides a programming interface for the multithread model. Notably, the multithread primitive database 1206 includes primitives for starting a thread, stopping a thread, suspending a thread, as well as synchronization of such starting, stopping, and suspending among threads. In addition, primitives are provided for indicating status information for individual threads, such as completion or suspension, among other threads. Furthermore, primitives may be provided for allowing data communication among threads.
A designer may interact with the input section 1202 to produce specification data for a multithreading system. The designer may work directly with the multithread primitive database 1206, or may work indirectly with the multithread primitive database 1206 through an alternative design tool defined within the input section 1202. The multithreading system specification data may be programmatic or may be interpretive (e.g., XML).
The multithread model section 1204 is coupled to the input section 1202 for receiving the multithreading system specification data. The multithread model section 1204 is also coupled to a database that stores the features or attributes of the multithread model (“multithread model database 1208”), and a database that stores the physical attributes of an FPGA (“FPGA database 1210”). The multithread model section 1204 includes a first section 1203 for generating a logical view of the multithreading system in accordance with the multithreading system specification. The logical view is defined in terms of the logical components of the multithread model stored in the multithread database 1208. The multithread model section 1204 includes a second portion 1205 for generating a physical view of the multithreading system based on the logical view. Notably, using information in the FPGA database 1210, the multithread model section 1204 maps the logical view of the multithreading system onto the physical components associated with an FPGA. The multithread model section 1204 provides FPGA design data as output.
Notably, the thread model 1300 may be physically implemented in programmable logic of an FPGA as a synchronous FSM. That is, a clock drives the state machine's transitions and, within each state of the state machine, operations are performed on operands producing outputs. In another embodiment, the thread model 1300 may be implemented using a microcontroller embedded within an FPGA. In yet another embodiment, the thread model 1300 may be implemented using a hard or soft microprocessor embedded within an FPGA.
The thread 1302 includes control logic 1320 for processing data and producing control state data 1316 and operation state data 1318. The control state data 1316 captures the state of the thread 1302 in terms of inter-thread communication (e.g., the thread is finished or suspended). The operation state data 1318 captures the internal state of the thread 1302, which is defined in accordance with the operation of the control logic 1320.
In one embodiment of the invention, the thread 1302 includes an IDLE control state that corresponds to the thread 1302 not currently carrying out any operation. During the IDLE control state, the isFinished terminal 1312 is asserted to indicate that the thread 1302 is not doing any work. If the start terminal 1304 is asserted, the thread 1302 moves out of the IDLE control state and performs the various operations that the thread 1302 is configured to perform. The isFinished terminal 1312 is no longer asserted to indicate that the thread 1302 is in operation. Asserting the stop terminal 1306 returns the thread 1302 to the IDLE control state, terminating all operations. Asserting the suspend terminal 1308 causes the thread 1302 to remain in its current operation state, regardless of how many clock cycles occur on the clock terminal 1310. The suspend terminal 1308 may be used during debugging or testing through an external agent. The isSuspended terminal 1314 is asserted while the thread 1302 is suspended.
The operation state data 1318 depends on the operational configuration of the control logic 1320 of the thread 1302 in accordance with specification data provided by a designer. That is, the control logic 1320 of the thread 1302 is configured to execute a series of steps, where one or more operations are performed at each step. In one embodiment, the thread 1302 may be designed to have one or more designated terminal operation states that cause the thread 1302 to enter the IDLE control state and assert the isFinished terminal 1312. Alternatively, there may be no such designated terminal control states, in which case the thread is control externally by asserting the stop signal.
For example, a VHDL code fragment for implementing the thread 1302 may be:
update: process (clk, stopThread)
begin—process update
The thread 1302 has thus far been described with respect to an external agent that asserts the start, stop, and suspend terminals. The “external agent” may be another thread, thus allowing threads to control other threads. In order to enable one thread to write or read the signals of another thread, the multithread model employs an interconnection topology and an addressing mechanism. That is, a plurality of threads are interconnected for communication amongst themselves, and a thread may associate the address of another thread with its collection of input and output signals.
Notably,
In one embodiment of the invention, the output bus 1408 of a thread model 1402 comprises a start control bus 1410, a stop control bus 1412, a suspend control bus 1414, and a status bus 1416. The width of each of the start control bus 1410 and the stop control bus 1412 is N (i.e., the number of thread models 1402 in the multithread model 1400). The width of the suspend control bus 1414 may be N−1 if self-suspension of a thread is not defined. The status bus 1416 may comprise M status signals (e.g., isFinished and isSuspended signals) and thus the width of the status bus 1416 is M. The input bus 1406 of a thread model 1406 comprises a start terminal 1422, a stop terminal 1424, and a suspend terminal 1426.
In one embodiment of the invention, each of the thread models 1402 produces a control signal for each of the other thread models 1402 (e.g., thread model 14021 produces control signals for thread models 14022 through 1402N) through operation of control logic 1430. Thus, if there are eight thread models 1402 in the multithread model 1400, then each thread model 1402 will produce eight start signals, eight stop signals, and seven suspend signals in accordance with the control logic 1430. For each of the thread models 1402, start signal and one stop signal is connected in a self-loop, which allows a thread model 1402 to start and stop itself. The interconnection component 1404 facilitates the connection between the output bus 1408 of a given thread model and the input buses of each of the other thread models.
Notably, if it is not intended that a particular one of the threads 1502 control the starting of the thread 15021, then the start control terminal 1508 is not asserted. If one of the threads 1502 intends to start the thread 15021, the start control terminal 1508 of that thread is asserted. For example, if the thread 15023 intends to start the thread 15021, then the thread 15023 asserts its start control terminal 1508. The start control terminals 1508 are logically OR-ed together by the OR component 1504 such that if one or more is asserted, the thread 15021 will be started. For purposes of clarity by example, a synchronization mechanism for starting the thread 15021 is shown. It is to be understood, however, that the multithread system 1500 may be extended to start, stop, and suspend any number of threads.
Returning to
In yet another embodiment, each of the thread models 1402 may include an identifier 1420 (e.g., a memory address). A pull mechanism may be employed to retrieve data from one thread model using another thread model. A thread model 1402 requiring data provides the identifier to the thread model 1402 in possession of the data, together with a read request, and the data item is provided after some latency. In yet another embodiment, a push mechanism may be employed, whereby once a thread model 1402 has data, it pushes the data to all threads that are known to require the data, together with its identifier 1420.
Programming Interface
In one embodiment of the invention, the set of primitives 1604 comprises FOP primitives 1606, COP primitives 1608, PIP primitives 1610, signal grouping primitives 1612, inter-process communication primitives 1614, memory element primitives 1616, run-time primitives 1618, implementation metric primitives 1620, and debugging primitives 1622. Each of the aforementioned primitives is discussed in detail below.
The FOP primitives 1606 provide a coding environment targeting multiple threads that operate in parallel. The FOP primitives 1606 include instruction set primitives 1624 and physical implementation primitives 1626. The instruction set primitives 1624 are used to program the threads. That is, the instruction set primitives 1624 provide a mechanism for establishing an instruction set of a thread, where the instruction set itself is programmable. Thus, a designer may modify an instruction set for a thread as desired (e.g., providing a domain-specific set of instructions). The physical implementation primitives 1626 are used to define the physical implementation of a given thread. For example, a thread may be implemented in programmable logic of an FPGA or in a hard or soft microprocessor or using a microcontroller.
In one embodiment, each thread is implemented as a custom FSM in programmable logic of an FPGA. An instruction set is defined for the FSM thread, where each instruction has a dedicated implementation. There is no additional support required for unused operations in the instruction set and multiple instructions may be executed simultaneously.
The COP primitives 1608 are used to include a function block into the design. In one embodiment, the COP primitives 1608 comprise “include” type primitives for specifying a particular function block to include within the design.
The PIP primitives 1610 may comprise instruction set primitives 1628 and physical implementation primitives 1630. The instruction set primitives 1628 are used to define an instruction set for a thread in a similar manner to the FOP primitives 1606 described above. Unlike the FOP primitives 1606, however, the instruction set primitives 1628 may be used to define certain system instructions. The system instructions are used to communicate with input/output interface logic blocks that communication with another system (within the FPGA or external thereto). For example, an interface logic block may be a gigabit Ethernet MAC core. The instruction set primitives 1628 provide support for different communication protocols to read/write data over various interfaces. For example, one type of interface may be completely streaming, with data arriving at every clock cycle. Another type of interface may have flow control, where there may be a pause in the data stream.
The physical implementation primitives 1630 define the physical implementation of the PIP (e.g., FSM, microprocessor). The PIP primitives 1610 may also comprise include type primitives for specifying the inclusion of the interface function block. Each interface block may have multiple ports, or groups of signals, associated therewith. One group of signals contains the connectivity to the external environment. The others connect to one or more PIP threads. For example, an interface block may have a set of signals that form a receive port and another set of signals that form a transmit port. In this case, the signals may be grouped together such that each port is assigned to a different PIP thread.
The signal grouping primitives 1612 are used to define signal groups. Grouping of signals may occur in various contexts, such as when connecting an interface block to a PIP thread, as described above, or when connecting to a memory element. In such cases, the programming information for an element such as a FOP thread states that the FOP thread is connected to another element.
The inter-process communication primitives 1614 provide support for synchronization and data communication between threads. Some basic aspects of the mechanism, such as support for starting, stopping, and suspending processes, may be built into the soft platform architecture. Thread synchronization and data communication, however, may be completely specified by a designer. In one embodiment, connections are explicitly specified between processes. Alternatively, required connections may be inferred from the operations defined for a particular group of processes. For example, an operation to start another process may have the form of “START(process)” or an operation to pass a data value to another process may have the form of “PASS(data, destination process).” With such an operation, a connection may be inferred without a designer explicitly defining the connection.
The memory element primitives 1616 are used to define the various types, sizes, and interconnections of memory elements. The memory element primitives 1616 may include the MIDL primitives discussed above in the section entitled “MEMORY MODEL,” for specifying the logical view of a memory subsystem.
The run-time primitives 1618 may be used to apply run-time reconfiguration. Run-time reconfiguration involves the modification of a circuit implemented within an FPGA at run-time. Dynamic reconfiguration for an FPGA is well-known in the art. For example, the run-time primitives 1614 may be used to migrate functionality between programmable logic and an embedded processor. Initially, some functionality is implemented in programmable logic with other functionality implemented using an embedded microprocessor. Implicitly, the programmable logic implementation exhibits higher performance than the processor implementation. Functionality may be offloaded to the processor to save area within the programmable logic. During execution, statistics may be taken to give feedback on the chosen partition. A reconfiguration controller determines a new partition and reconfigures the FPGA. An exemplary decision condition would be based on the frequency of events. More frequency events may thus be handled in programmable logic, with less frequency events handled by the embedded processor.
The implementation metric primitives 1620 may be used to define the requirements of the system. For example, the implementation metric primitives 1620 may be used to establish performance requirements that must be met. The implementation metric primitives 1620 may be used to create constraints files (e.g., timing constraint files) that can be used by FPGA design tools (e.g., map/place/route tools). The implementation metric primitives may also provide low-level optimizations (e.g., clock frequency requirements, throughput latency requirements), as well as high-level optimizations (e.g., optimize for area, throughput, latency, power, and the like).
The debugging primitives 1622 may be used to provide debugging capabilities. The debugging primitives 1622 may be used to capture simulated data associated with the soft platform architecture. For example, instead of presenting signal waveforms to the designer, the debugging primitives 1622 allow for data presentation in a more abstracted form. The debugging primitives 1622 also provide lower-level functions through the use of tags or commands that cause the circuitry to be modified and operate in a debug mode.
Appendix C shows exemplary XML code that defines interface logic for a logic block or “core,” referred to as “Aurora,” which may be used to drive the soft platform architecture described herein. The Aurora interface, available from Xilinx® of San Jose, Calif., is implemented as external intellectual property for point-to-point communication over multi-gigabit transceivers. The first set of signals (clk, reset, RXN, RXP, TXN, and TXP) represent the signals that connect to the external environment. In this example, the data lines would be tied to serial transceivers. The port labeled “rx” is the receive port and has several signals associated therewith. Similarly, the port labeled “tx” is the transmit port and also has several signals associated therewith.
Within each port is a clock associated therewith. The clock determines the clock domain. In the present example, both “rx” and “tx” ports have an output clock. Thus, the IP function block has circuitry to generate a clock signal. This clock would drive all threads in the determined clock domain. Alternatively, a “useclk” tag may be used if the IP block does not generate a clock signal. The clock that drives the port is also used to drive other threads in the clock domain.
Appendix D illustrates an exemplary XML code of a PIP thread that handles the receive-side connection to the Aurora interface defined in Appendix C. The PIP thread reads data from the receive port of the Aurora interface and stores the data in a buffer. The protocol for the interface includes flags marking the state of a frame, end of frame, and whether data is valid. The data valid (RXSourceReadyBar) signal allows the stream to pause. The PIP thread waits until the entire frame has been received before committing the frame to memory. Committing the frame to memory is an indicating mechanism informing the buffer that an entire frame is in memory. This ensures that other blocks of logic that read from the memory do not process a partial frame.
Line 2 of Appendix D illustrates that the PIP thread connects to the A port of a memory named “a2e_buf.” Line 3 of Appendix D shows that the PIP thread also connects to the rx port of the interface block named Aurora (as shown in Appendix C). Lines 4-8 of Appendix D define the variables of the PIP thread. In the present example, only internal variables are shown, but variables may also be defined to connect to other threads.
Appendix E illustrates exemplary XML code for effecting an explicit connection. A thread named “sender” having an output named “myout” is defined. Threads named “receiver—1” and “receiver—2”, each with an input named “myin,” are also defined. The defined connection will connect the output of the sender thread with the two input ports of the receiver threads.
While the foregoing describes exemplary embodiment(s) in accordance with one or more aspects of the present invention, other and further embodiment(s) in accordance with the one or more aspects of the present invention may be devised without departing from the scope thereof, which is determined by the claim(s) that follow and equivalents thereof. Claim(s) listing steps do not imply any order of the steps. Trademarks are the property of their respective owners.
MEMORY top
{
ARCHITECTURE
{
}
MEMINTERFACE
{
}
MEMICINTERFACE
{
}
MEMORY Mem1
{
MEMORY Mem2
{
}
}
entity IPv4_handler is
generic (
TID: integer:=IPv4_HANDLER_TID);
port (
clk: in std_logic;
Control for this Thread
startThread: in std_logic;
stopThread: in std_logic;
suspendThread: in std_logic;
Status for this Thread
threadIsBlocked: out std_logic;
threadIsFinished: out std_logic;
Control and Status for Other Threads
isFinished: in std_logic_vector((NUM_THREADS—1)
downto 0);
start: out std_logic_vector((NUM_THREADS—1)
downto 0);
stop: out std_logic_vector((NUM_THREADS—1)
downto 0);
suspend: out std_logic_vector((NUM_THREADS—2) downto 0);
is Blocked: in std_logic_vector((NUM_THREADS—1)
downto 0));
<hook name=“aurora”>
<connection name=“c1”>
| Number | Name | Date | Kind |
|---|---|---|---|
| 5113392 | Takiyasu et al. | May 1992 | A |
| 5778230 | Wimble et al. | Jul 1998 | A |
| 5867180 | Katayama et al. | Feb 1999 | A |
| 6078736 | Guccione | Jun 2000 | A |
| 6182183 | Wingard et al. | Jan 2001 | B1 |
| 6230307 | Davis et al. | May 2001 | B1 |
| 6324629 | Kulkarni et al. | Nov 2001 | B1 |
| 6405160 | Djaja et al. | Jun 2002 | B1 |
| 6408163 | Fik | Jun 2002 | B1 |
| 6581187 | Gupta et al. | Jun 2003 | B2 |
| 6647431 | Utas | Nov 2003 | B1 |
| 6671869 | Davidson et al. | Dec 2003 | B2 |
| 6704914 | Nishida et al. | Mar 2004 | B2 |
| 6735770 | Yeager et al. | May 2004 | B1 |
| 6794896 | Brebner | Sep 2004 | B1 |
| 6891397 | Brebner | May 2005 | B1 |
| 6904431 | Holmgren | Jun 2005 | B2 |
| 6918103 | Brawn et al. | Jul 2005 | B2 |
| 6952825 | Cockx et al. | Oct 2005 | B1 |
| 6983461 | Hutchison et al. | Jan 2006 | B2 |
| 7058562 | Powell | Jun 2006 | B2 |
| 7131077 | James-Roxby et al. | Oct 2006 | B1 |
| 7137116 | Parkes et al. | Nov 2006 | B2 |
| 7185309 | Kulkarni et al. | Feb 2007 | B1 |
| 7209979 | Solomon et al. | Apr 2007 | B2 |
| 7228520 | Keller et al. | Jun 2007 | B1 |
| 7376954 | Kissell | May 2008 | B2 |
| 7379451 | Brebner | May 2008 | B1 |
| 7401333 | Vandeweerd | Jul 2008 | B2 |
| 20020067717 | Raschke et al. | Jun 2002 | A1 |
| 20020120912 | He et al. | Aug 2002 | A1 |
| 20020129173 | Weber et al. | Sep 2002 | A1 |
| 20020162097 | Meribout | Oct 2002 | A1 |
| 20020170038 | Yeh et al. | Nov 2002 | A1 |
| 20020188923 | Ohnishi | Dec 2002 | A1 |
| 20030028854 | Nishida et al. | Feb 2003 | A1 |
| 20030033374 | Horn et al. | Feb 2003 | A1 |
| 20030056084 | Holgate et al. | Mar 2003 | A1 |
| 20030097481 | Richter | May 2003 | A1 |
| 20030126186 | Rodgers et al. | Jul 2003 | A1 |
| 20030126195 | Reynolds et al. | Jul 2003 | A1 |
| 20030144980 | Holmgren | Jul 2003 | A1 |
| 20030145210 | Taylor | Jul 2003 | A1 |
| 20030182083 | Schwenke et al. | Sep 2003 | A1 |
| 20030188099 | Baxter, III | Oct 2003 | A1 |
| 20030188100 | Solomon et al. | Oct 2003 | A1 |
| 20040006584 | Vandeweerd | Jan 2004 | A1 |
| 20040071152 | Wolrich et al. | Apr 2004 | A1 |
| 20040078448 | Malik et al. | Apr 2004 | A1 |
| 20040083269 | Cummins | Apr 2004 | A1 |
| 20040128120 | Coburn et al. | Jul 2004 | A1 |
| 20040187112 | Potter | Sep 2004 | A1 |
| 20040225965 | Garside et al. | Nov 2004 | A1 |
| 20050034090 | Sato et al. | Feb 2005 | A1 |
| 20050038806 | Ma | Feb 2005 | A1 |
| 20050086565 | Thompson et al. | Apr 2005 | A1 |
| 20050114593 | Cassell et al. | May 2005 | A1 |
| 20050172085 | Klingman | Aug 2005 | A1 |
| 20050172087 | Klingman | Aug 2005 | A1 |
| 20050172088 | Klingman | Aug 2005 | A1 |
| 20050172089 | Klingman | Aug 2005 | A1 |
| 20050172090 | Klingman | Aug 2005 | A1 |
| 20050172289 | Klingman | Aug 2005 | A1 |
| 20050172290 | Klingman | Aug 2005 | A1 |
| 20050177671 | Klingman | Aug 2005 | A1 |
| 20050210178 | Klingman | Sep 2005 | A1 |
| 20050223384 | Klingman | Oct 2005 | A1 |
| 20050262286 | Klingman | Nov 2005 | A1 |
| 20060041684 | Daniell et al. | Feb 2006 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10769592 | Jan 2004 | US |
| Child | 12465247 | US |
